Ten of the assistant's answers come from live tools rather than the model. Give it the concrete input in the middle column and it fetches real figures.
| Ask for | Give it | You get back |
|---|
| What ranks right now | a search query | the live Google results page: ranked results, plus any answer box, People Also Ask, knowledge panel or AI Overview showing for it |
| Where you rank | a keyword and a domain | the domain's position in that live result set, or an honest "not in the top 10" |
| What people type | a partial query | Google's live autocomplete suggestions for it |
| Keyword ideas | a seed keyword | up to 25 related keywords with search volume, competition, CPC, difficulty, and intent |
| A site's keyword footprint | a domain | the keywords it currently ranks for, with position, volume, and landing URL |
| AI Overview visibility | a domain | keywords where that domain already appears in Google AI Overviews, with volume and rank |
| How strong a site is | a domain | its Ahrefs Domain Rating, a 0–100 read on backlink authority |
| Link profile shape | a domain | total backlinks, referring domains, the dofollow/nofollow split, referring IPs, broken links, spam score |
| Traffic and value | a domain | estimated monthly organic visits, ranking keyword count, traffic value, and the position spread |
| Who you compete with | a domain | the domains ranking for the same keywords, with shared-keyword counts |
Prompts that work well:
- "What's ranking for best crm for startups right now, and what does the SERP want?"
- "Is example.com ranking for seo audit tool?"
- "Find keyword ideas for project management software and show search volume and difficulty."
- "Summarise the backlink profile and organic traffic of ahrefs.com."
- "Which keywords does notion.so already rank for in Google AI Overviews?"
Five workflows are packaged as one-tap commands. Type / in the message box, or tap a card, and the assistant loads a full playbook before it starts: /serp-analysis, /rank-check, /keyword-research, /competitor-teardown, and /ai-visibility-audit. A teardown, for example, resolves the brand, then pulls authority, traffic, ranking keywords and link profile, then reports three exploitable gaps rather than four tables.
It chains tools up to eight hops in a single turn. Give it a brand name and nothing else and it resolves the name to a domain, then pulls what it needs before it answers, rather than asking you for the URL first.
Ask in German, or about a .de domain, and it switches markets. Keyword and SERP lookups run against either the United States or Germany, and the assistant names which market a number came from.
Everything else, from content structure to on-page fixes to how to earn a citation, it answers from the model directly. When a tool fails or a query returns nothing, it tells you, rather than inventing a number.